Ask about what people did, not what they would do. "Tell me about the last time you rescheduled a lesson" gets facts. "Would you use an app for rescheduling?" gets a polite yes. Build every question around a real past event, and never pitch your idea until the end.

This is for founders who want to talk to customers before they build. The script and examples follow one case: an app for independent piano teachers who juggle lesson times and payments by text message.

See it in actionAsk about the last time, not the future
“Would you use an app for rescheduling?” gets a polite yes. “Tell me about the last time a lesson had to move” gets facts you can use. Illustrative example. Use the playback controls or the thin timeline at the top to explore.

In short:

  • Predictions about future use are unreliable. Past behaviour is the evidence.
  • Seven questions fit in 20 minutes and cover the full story.
  • Leading questions hide the answer inside the question. Rewrite them as open prompts.
  • AI is good at checking your script and sorting notes, not at asking.

Past behaviour is the only answer you can trust

People are bad at predicting their own behaviour. They are good at describing what happened last Tuesday. So a strong interview asks for stories, with dates, tools and costs.

Rob Fitzpatrick's book The Mom Test puts it in three rules. Talk about their life, not your idea. Ask about specifics in the past, not opinions about the future. Talk less and listen more.

Y Combinator's talk How to talk to users gives similar advice from the startup side. A useful rule of thumb: if someone hasn't tried to solve the problem, it probably isn't painful enough to pay for.

A 20-minute script with seven questions

Use these in order. Each follows the last story, so the conversation feels natural.

  1. Set the scene. "How do you run your lessons today, from booking to getting paid?"
  2. Find a recent event. "Tell me about the last time a lesson had to move."
  3. Get the detail. "What did you do, step by step? Which apps or messages did you use?"
  4. Measure the cost and frequency. "How long did that take? How often does it happen in a normal month?"
  5. Find past attempts. "Have you tried anything to fix this? What happened?"
  6. Find the spend. "Do you pay for any tool for this today? What made you choose it?"
  7. Ask for the next person. "Who else should I talk to about this?"

Question 5 matters most. A teacher who tried two calendar apps and gave up has a real problem. A teacher who never looked has a mild one.

Question 7 keeps the pipeline full. Good interviews lead to more interviews.

Keep the idea out until the last two minutes

If you describe your app early, every answer bends towards it. People want to be kind. Save the pitch for the end, and even then, watch for actions rather than praise. "Can I try it when it's ready?" with an email is a signal. "Sounds great" is not.

Leading questions put the answer in your mouth

A leading question suggests the answer you hope for. The Nielsen Norman Group's guide to avoiding leading questions shows how small wording changes skew what people say.

Every leading question has a neutral version that asks for a story instead.

Leading question Neutral version
"Isn't scheduling a pain?" "How do you handle scheduling?"
"Would you pay for this?" "What do you pay for today?"
"Do you hate chasing payments?" "Tell me about the last late payment."
"Would an app help?" "What have you tried so far?"
"How much time would it save?" "How long did it take last time?"

Two more traps to avoid:

  • Yes or no questions end the story. Start with "how", "what" or "tell me about". NN/g's guide to open-ended questions has more examples.
  • Hypotheticals like "if there was an app that..." invite fiction. Return to the last real event.

Silence gets you the best answers

After someone answers, wait three seconds before you speak. People often add the most useful detail after a pause: "Actually, the worst part is when two parents both think they booked Thursday."

Also, ask "why" gently and only once or twice per topic. "What made that hard?" sounds curious. A string of "why?" sounds like a test.

AI can prepare and sort, but you ask

AI assistants are useful before and after the call. They should not replace the call.

Before: paste your script into an assistant and ask it to flag leading, yes or no, and hypothetical questions. It catches many of them in seconds.

After: if the person agreed to recording, a transcript lets an assistant group answers into themes. Ask for direct quotes with each theme, and check them against the transcript. Ask first, store recordings carefully, and delete them when you're done.

A good prompt for sorting notes:

Here are notes from 6 interviews with piano teachers.
              Group what they said into problems. For each problem:
              - how many of the 6 mentioned it unprompted
              - one exact quote per person who mentioned it
              - what they use today
              Do not add problems that are not in the notes.

The last line matters. Without it, assistants can add plausible problems nobody mentioned.

Five to eight interviews show the pattern

You don't need a big sample to hear whether a problem is real. Keep going until the last two interviews told you nothing new. For one tight customer group, that is often five to eight conversations.

If answers are all over the place, the group is too broad. "Music teachers" might split into teachers with 5 students and schools with 50. Pick one and interview again.

To find people, start with your own network, one step removed. For paid recruiting, panels such as User Interviews find participants who match a profile. A scheduling link from a tool like Calendly saves a round of emails.

Bring the interviews into GrowApps

When you start a new app in GrowApps, it first asks a few questions: your relationship to the idea, how the problem is handled today, and what a win looks like. Answers taken from real interviews make those answers specific, for example "teachers lose about one lesson payment a month to confusion".

GrowApps then researches apps like yours and shows screen designs for approval before it writes code, so interview findings shape the first version (how GrowApps works).

Check your own script in ten minutes

Write your five most important questions. Paste them into any AI assistant with this line: "Flag every question that is leading, yes or no, or hypothetical, and rewrite it to ask about a past event." Fix what it flags, then book your first call. Next, see how to validate an app idea in a weekend.

Frequently asked questions

Should I pay people for an interview?

For hard-to-reach professionals, a small gift card or donation helps. For people with a strong problem, the chance to shape a tool is often enough. Paid strangers from panels are fine, but screen them carefully so they really fit your customer.

Can I use a survey instead?

Surveys count how common something is. Interviews find out what is going on. Do interviews first, then write a validation survey to measure the problems you heard.

Sources and further reading